Title: Telecommunications
1Telecommunications Networks Lab
Mesh Networking
Angelakis Vangelis
20 / 1 / 2005
2Introduction
In the past decade the Ad hoc Networking
paradigm absorbed a lot of research
effort. Most of the work is focused on isolated
military or specialized civilian application
multihop ad hoc networks. Turning ad-hoc
networks into a commodity takes a few
changes Make multihop flexible low cost last
mile extensions of wired infrastructure Turn
them into MESH NETWORKS!
Angelakis Vangelis
20/1/2005
3Presentation Layout
Wireless Mobile Ad Hoc Networks (MANETs)
Criticism Wireless Mesh Networks MANET WMN
common grounds Standards Committees Research
problems Hyacinth Routing Metrics Locustworld
Angelakis Vangelis
20/1/2005
4Where is the ad-hoc paradigm failing?
An ad-hoc network is a collection of mobile
nodes that connect over the wireless medium
without the need of any pre-deployed
infrastructure. Nodes in a MANET can dynamically
self-organize into temporary and arbitrary and
network topologies No no pre-existing
infrastructure required Supporting Scenarios
Disaster Recovery Areas Battlefields - Network
Isolation
Angelakis Vangelis
20/1/2005
5Where is the ad-hoc paradigm failing?
- Key Research drivers
- Bluetooth (802.15.5) and 802.11 mass market
deployment - IETF MANET WG standardizations
- from the USA DoD
- Main problem
- Users want
- affordable devices
- Internet access
- Is high quality connectivity during mobility such
an important issue?
Angelakis Vangelis
20/1/2005
6Making ad-hoc networks a commodity
- Pass down as much research from the MANET field
to a more market-viable networking paradigm. - Relax the main constraint of MANETS
- Accept the existence of infrastructure.
- Wireless Mesh Networks (WMNs) are build by
interconnecting internet egress points with
end-user devices can act as terminals and as
routers. - Community Networks being the earliest form of
Meshes, relaxed a second MANET feature - Mesh nodes are practically stationary
Angelakis Vangelis
20/1/2005
7How does this change look like?
A 10-node MANET at time t0
Angelakis Vangelis
20/1/2005
8How does this change look like?
A 10-node MANET at time t0
Angelakis Vangelis
20/1/2005
9How does this change look like?
A 10-node MANET at time t1
Angelakis Vangelis
20/1/2005
10How does this change look like?
A wireless mesh network of 3 tiers
Wired Backbone
Angelakis Vangelis
20/1/2005
11Ad-hoc WMN, what is different
- The WMN concept is similar to the one of ad-hoc
networks. - With four important differences
- (practically) fixed nodes gt Topology changes are
infrequent - Addition of nodes
- Node failure or maintenance
- Traffic distribution is skewed (to/from the wired
network) - Traffic characteristics aggregated from large
numbers of flows gt network optimization based on
profiling - Reactive discovery of multihop paths is not
efficient for an effective backbone and not fit
under (1).
Angelakis Vangelis
20/1/2005
12A research decision that made a big difference
- Testbeds were used in WMN research from the
start. - MIT Roofnet (Chambers 02)
- Proof of existence of good enough solution for
civilian applications has stimulated the users
interest to adopt this technology. - Two main solutions classes
- off-the-shelf ? community networks
- - Roofnet - CUWiN
- - BAWUG - Seattle Wireless
- proprietary ? MeshNetworks, Tropos Networks,
Radiant Networks, BelAir, Strix etc
Angelakis Vangelis
20/1/2005
13Standards Committees
- IEEE standards groups actively working to define
specs. for wireless mesh networking techniques - Special groups established to define the
requirements for mesh networking in WPANs, WLANs,
and WMANs. - The following standards amendments are expected
- 802.15.5 (bluetooth - WPAN)
- 802.11s (wi-fi - WLAN)
- 802.16a (wi-max - WMAN)
- Also 802.20 (wireless mobile broadband access
-WMBA) is to support the Mesh Networking paradigm
from the first spec. - (expected first draft sometime in the 2nd
semester of 2006)
Angelakis Vangelis
20/1/2005
14802.11s
The 802.11 group has set up the TG s to
discuss proposals for a specs amendment in the
2nd quarter of 05 and reach a final document no
sooner than 2007. Scope of the Project An
IEEE 802.11 Extended Service Set (ESS) Mesh is a
collection of APs interconnected with wireless
links that enable automatic topology learning and
dynamic path configuration. The proposed
amendment shall be an extension to the IEEE
802.11 MAC. The amendment will define an
architecture and protocol for providing an IEEE
802.11 ESS Mesh using the IEEE 802.11 MAC to
create an IEEE 802.11 Wireless Distribution
System that supports both broadcast/multicast and
unicast delivery at the MAC layer using
radio-aware metrics over self-configuring
multi-hop topologies. An ESS Mesh is functionally
equivalent to a wired ESS, with respect to the
STAs relationship with the BSS and ESS.
Angelakis Vangelis
20/1/2005
15802.11s
The amendment shall enable interoperable
formation and operation of an ESS Mesh, but shall
be extensible to allow for alternative path
selection metrics and/or protocols based on
application requirements. A target configuration
is up to 32 devices participating as AP
forwarders in the ESS Mesh. However, larger
configurations may also be contemplated by the
standard. It is intended that the architecture
defined by the amendment shall allow an ESS Mesh
to interface with higher layers and to connect
with other networks using higher layer
protocols. The amendment shall utilize IEEE
802.11i security mechanisms, or an extension
thereof, for the purpose of securing an ESS Mesh
in which all of the APs are controlled by a
single logical administrative entity for
security. The amendment shall allow the use of
one or more IEEE 802.11 radios on each AP in the
ESS Mesh.
Angelakis Vangelis
20/1/2005
16802.16a
- 802.16 is a P-MP first-mile/last-mile WMAN
connection standard. (data rate up to 120Mbps _at_
30miles). - A base station (BS) serves a number of subscriber
stations (SS). - BS uses a broadcast channel to transmit to all
SSs. - 802.16 approved in 2001 (10-66GHz operation
TDMA, TDDFDD) - 802.16a approved in Jan. 2003 (2-11 GHz operation
added -ODFM) - already obsolete and part of the 802.16-2004
doc - Stations may have direct links to each other -
control can be distributed. - WiMax forum formed later in 2003 (just like the
802.11 Wi-Fi forum) to promote IEEE standards for
interoperability) - No WiMax Forum Certified in the market yet.
Angelakis Vangelis
20/1/2005
17Major research activity by Microsoft
- One of the most significant research efforts in
the WMN field is conducted by the networking
research group of Microsoft - In June Microsoft organized the Mesh Networking
Summit 2004 - Gathering most of the major players from the
industry and academia community this was the
first focused event on WMNs - A presentation a few days later by Victor Bahl
sets the problem space and the wish list of WMNs
as has been formed after that meeting
Angelakis Vangelis
20/1/2005
18Problem space
- Range and Capacity
- Inexpensive electronically steerable directional
antenna or MIMO for range enhancement - Multiple frequency meshes
- Multi-radio hardware for capacity enhancement via
greater spectrum utilization - New data channel MAC with Interference management
or higher throughput - Multihop Routing
- L2.5 on-demand source routing with link quality
based routes selection - Route selection with multiple radios (multiple
channels) - Security, Privacy, and Fairness
- Guard against malicious users (and freeloaders)
- EAP-TLS between MeshBoxes, PEAPv2 or EAP-TLS
between clients and MeshBoxes - Priority based admission control, Secure
traceroute - Self Management Self Healing
- Minimal human intervention - avoid network
operator - Watchdog mechanism with data cleaning and liar
detection - Online simulation based fault isolation and
diagnosis
Angelakis Vangelis
20/1/2005
19Problem space II
- Smart Spectrum Utilization
- Spectrum etiquettes and/or rules
- Agile radios, cognitive radios, 60 GHz radio,
underlay technologies - Cognitive software applications
- Analytical Tools
- Information theoretic tools that predict network
viability performance with practical
constraints, based on experimental data - Ease of use (Plug and play, HCI)
- Pleasant, hassle-free user experience
- QoS protocols to improve content delivery
- Digital Rights Management (DRM)
- Broadband access popularity related to expanded
digital content. - Increase the value proposition for
end-users/subscribers
Angelakis Vangelis
20/1/2005
20Capacity Scalability
- Use of multiple radios per node
- channel assignment (self configuration)
- - neighbor to interface binding
- - interface to channel binding
- Problem Constrains
- Capacity of a radio channel within an
interference zone is limited - Neighbors that need to communicate must share
a common channel - of available channels per node of
available Wireless NICs - Fixed number of non-overlapping channels
- Dynamic channel assignment depends on link load.
So this problem becomes routing dependent
Angelakis Vangelis
20/1/2005
21Hyacinth
Distributed load balancing routing Channel
assignment Requires (at least) 2 wireless NIC
per node in order to avoid ripple effects when
switching channels Logically each wired gateway
is the root of a spanning tree and every node
attempts to participate in at least one such
spanning tree. Tree construction process similar
to 802.1d STP Each node who can reach the wired
network sends an ADVERTISE packet to its
neighbors that carries the cost this node has to
reach the wired network.
Angelakis Vangelis
20/1/2005
22Hyacinth
- Cost in Hyacinth is taken to be
- hop count (from advertising node to wired
gateway) - gateway link capacity (g/w uplink residual
capacity) - path capacity (minimum path residual capacity)
- Given the ADVERTISE packet each recipient chooses
to JOIN the sender or ignore it. - X receiving a JOIN packet from Y
- Forwards the request upward in the tree and
waits for an ACCEPT message - adds Y to its children list and
- Sends Y an ACCEPT message with channel and IP
information
Angelakis Vangelis
20/1/2005
23Hyacinth
Ripple effect problem
Angelakis Vangelis
20/1/2005
24Hyacinth
Neighbor-to-interface binding Each
node is responsible to assign channels to its
DOWN-NICs To ensure relay capability to the
wired medium nodes closer to it have higher
priority on channel selection
Angelakis Vangelis
20/1/2005
25Hyacinth
Interface to channel binding To assign a
channel to a DOWN-NIC a node need to be able to
estimate the usage status of all channels in its
interference neighborhood. Exchange of a
CHNL_USAGE message to all k1-hop neighbors
carrying load information on used channels. k
interference range A node with a loaded channel
calculates the total load of other channels based
on the contributed information to select a less
loaded channel in its interference range.
Angelakis Vangelis
20/1/2005
26Metrics
Hyacinth is a higher-to-lower cross layer
system Increase in capacity is attempted by
changing channels using routing and traffic load
information The other way around is to decide
on the routes based on physical channel
characteristics Selection of routes based on
existing link quality How? Take a well studied
ad-hoc routing protocol like DSR and do what we
tried last year Incorporate link information in
the Route Discovery phase packet exchange. Add a
metric maintenance/update mechanism.
Angelakis Vangelis
20/1/2005
27Link metrics
- Hop count
- Per-hop RTT
- Per-hop Packet Pair Delay
- Expected Transmission Count (ETX) MIT
- Weighted Cumulative Expected Transmission Time
(WCETT) -Microsoft
Angelakis Vangelis
20/1/2005
28Link metrics
1. Hop count Routing goal Minimize hop count
on route. Simplicity 2. Per hop RTT Send a
probe and get a probe ack. Routing goal
Minimize the total RTT sum. avoids highly
loaded or lossy links - queuing delay 3.
Per-hop Packet Pair Delay Send a pair of
probes A small one and then a large and measure
delay between them at receiving node. Routing
goal Minimize delay link data rate into
account - overhead
Angelakis Vangelis
20/1/2005
29Link metrics
4. ETX Estimate the number of retransmissions
required to send unicast packets by measuring
loss rate of broadcast packets. Routing Goal
minimize sum of expected retransmissions along a
route. Broadcast every second a probe packet
containing count of received probes from all
neighbors in the past 10 secs and use that to
calculate the probability of a correct
transmission.
Angelakis Vangelis
20/1/2005
30Link metrics
Eg. probability of a correct transmission A has
a packet to send to B Packet (A-gtB) AND
Acknowledgement (B-gtA) need be correct to avoid
retransmission. A received 8 probes from B in
the past 10 secs B reported receiving 9 probes
from A Assuming independent losses then
probability of single correct transmission is
A?B B?A
(1-0.1) x (1-0.2)0.72 Expected
nr. of retransmissions is 1/0.72 1.39
Angelakis Vangelis
20/1/2005
31Link metrics
- ETX pros and cons
- Probing overhead is small
- no delays measured so no self-interference
- broadcasts sent at 1Mbps 802.11b and 6 Mbps
802.11g so loss rate will be less than that of
unicast traffic rate - no link load or throughput correlation
Angelakis Vangelis
20/1/2005
32Link metrics
5. WCETT the first multi-radio metric With ETX
in mind WCETT is a path metric taking into
account the time it would take for a packet of
some fixed size S to go over each link of the
path. Routing Goal WCETT minimization Assume
ETT is known. ETT can bebased on the ETX
calculated previously ETT ETX x S /
B Forgetting multi-ratio, for an n-hop path
Angelakis Vangelis
20/1/2005
33Link metrics
Assume that two co-channel hops on a path always
interfere with eachother , then defining Xj to
be the sum of transmission times on a channel j,
for each channel k in the system one can
argue that the total path throughput will be
dominated by the bottleneck channel so
Actual metric
Angelakis Vangelis
20/1/2005
34Link metrics
- From a design viewpoint WCETT achieves 3 goals
- Takes loss rate bandwidth into account (ETT
definition) - Is increasing over the path length
- Accounts for throughput reduction due to
co-channel - interference of links on a path.
Angelakis Vangelis
20/1/2005
35locustworld
A diverse sellable mesh-in-a-box
solution Linux modified kernel
boot-from-cd and ready to play Meshing
operation based on a good mingling of known
protocols
Angelakis Vangelis
20/1/2005
36locustworld
The software boots, it then allocates itself an
address, typically in the 10.x.x.x range.
Initially the allocated address is random.
Then, it attempts to find an internet gateway,
first probing 192.168.1.1/255.255.128.0 and then
using a DHCP client on the Ethernet interface to
try to sign on to a gateway. If no gateway can be
found, the software considers that it has only
wireless links and is a repeater-cell. The cell
starts an internal dns server and transparent web
proxy on port 80 and 8080, a dhcp server is also
started an a random class C network is picked in
the range 192.168.128.0/255.255.128.0 Clients
connecting to the dhcp cell are pointed to the
wireless interface for default gateway and dns
server. The dns server running on the repeater
always returns the address of the gateway, no
matter what domain name is resolved. At this
point, the node still cannot serve clients, but
it will sign them on. Meanwhile, an AODV module
is loaded and the node then finds neighbour cells
in its local range by sending/receiving UDP
packets to the broadcast address.
Angelakis Vangelis
20/1/2005
37locustworld
Any cells within the mesh which are gateways,
periodically broadcast a route to a bogus address
which implies an internet gateway. This route is
repeated outwards in a ring fashion as per the
AODV protocol. Any cell without a gateway
receiving this address then attempts to establish
a compressed encrypted IP-tunnel VPN via the
vtund package. This IP tunnel could be over
multiple hops to the destination gateway, AODV
handles the optimised routing between linked
cells. The cell then switches all its
outbound dns and ip traffic to go via this VPN
gateway link. The DHCP configuration is also
updated to now serve the remote gateway address
as a dns server (gateway nodes run a real dns
proxy) Any clients who signed on before the
link was found will forward traffic to the local
cell which will proxy it via http proxy etc, any
clients signing on after a gateway is found will
receive the remote gateway details and will have
full IP routing. Any client signing directly on
to a cell which has a local internet gateway will
go directly via that gateway. Generally, all
end user clients, eg non-mesh are routed via
standard NAT
Angelakis Vangelis
20/1/2005
38Further Reading Some Pointers
Jinyang Li, et.al., Capacity of Ad Hoc Wireless
Networks, (MobiCom '01), Rome, July 2001. B. A.
Chambers The Grid Roofnet a Rooftop Ad Hoc
Wireless Network'' MS Thesis, Dept of
Electrical Engg and Computer Science,
MIT. Hyacinth An IEEE 802.11-based
Multi-channel Wireless Mesh Network
http//www.ecsl.cs.sunysb.edu/multichannel/ R.
Bruno, M. Conti, E. Gregori Mesh Networks
Commodity Multi-hop Ad-Hoc Networks R. Draves,
et. al. Comparison of Routing Metrics for Static
Multi-Hop Wireless Networks ACM SIGCOMM,
Portland, OR, August 2004. R. Draves, et. al.
Routing in Multi-radio, Multi-hop Wireless Mesh
NetworksACM MobiCom, Philadelphia, PA, September
2004.
TNL - Mobile Computing Group Angelakis
Vangelis 21/10/2003
39Further Reading Some Pointers
- http//research.microsoft.com/mesh
- Roofnet
- Seattle Wireless
- Locust World
- Kingsbride Link
- Mesh Networks Inc
- FireTide Inc.
- Strix Networks Inc
- Telabria Inc
TNL - Mobile Computing Group Angelakis
Vangelis 21/10/2003